A hybrid SARIMA wavelet framework for groundwater level forecasting and managed aquifer recharge assessment in semi arid regions: Application to the Galedar Aquifer, southern Iran

Groundwater depletion in semi‑arid regions is a global crisis demanding accurate forecasting and proactive intervention. This study develops a transferable hybrid SARIMA‑wavelet framework to forecast water levels (WLs) and evaluate managed aquifer recharge (MAR) scenarios. The framework is demonstrated on the critically stressed Galedar aquifer (southern Iran, 2006‑2024) as a test case. Wavelet decomposition reveals that decadal variability dominates regional flow zones (48.5% of variance), while annual variability dominates high‑recovery zones (45%). Manual Box‑Jenkins SARIMA models, built with seasonal differencing (D=1), achieve out‑of‑sample RMSE of 0.062–0.162 m. Sensitivity analysis identifies specific yield (Sy) as the most influential parameter globally (Sobol’ ST = 0.58–0.71). The hybrid model reduces RMSE by 16‑31% compared to standard SARIMA, MODFLOW, and LSTM. Under business‑as‑usual (BAU), the mean WL declines 2.6 m (0.26 m/year) by 2034. MAR scenarios show: high‑recovery zone recharge (+1.41 m, 71% decline reduction); distributed recharge (1.8‑fold greater storage recovery per unit recharge); episodic recharge (+1.11 m, 40% lower efficiency). Cost‑benefit analysis confirms positive net present value (BCR 1.4‑3.4). The proposed framework is generalizable to any overexploited aquifer in arid/semi‑arid regions and supports risk‑based, spatially stratified management (SDG 6 & 13).半干旱地区的地下水枯竭是一场全球性危机，亟需开展精准预测与主动干预。本研究构建了一种可迁移的混合SARIMA-小波框架，用于预测水位（Water Levels, WLs）并评估含水层人工补给（Managed Aquifer Recharge, MAR）情景。本研究以承压严重的盖勒达尔含水层（伊朗南部，2006-2024年）为测试案例，对该框架进行了验证。小波分解结果显示，区域流动带的方差主导因素为年代际变率（占总方差的48.5%），而高补给带的方差主导因素为年际变率（占45%）。采用季节差分（D=1）构建的手动博克斯-詹金斯SARIMA模型，其样本外均方根误差（Root Mean Square Error, RMSE）为0.062~0.162米。敏感性分析结果表明，比产率（Specific Yield, Sy）是全球范围内影响最大的参数（索伯尔总阶灵敏度指数Sobol’ ST=0.58~0.71）。与标准SARIMA、MODFLOW以及长短期记忆网络（Long Short-Term Memory, LSTM）相比，该混合模型的RMSE降低了16%~31%。在常规情景（Business-as-Usual, BAU）下，到2034年平均水位将下降2.6米（年均下降0.26米）。含水层人工补给情景结果如下：高补给带补给（水位提升1.41米，降幅减少71%）；分布式补给（单位补给量的储层恢复量提升1.8倍）；间歇式补给（水位提升1.11米，效率降低40%）。成本效益分析证实，该方案具有正净现值，效益成本比（Benefit-Cost Ratio, BCR）为1.4~3.4。本研究提出的框架可推广至干旱/半干旱地区任何超采含水层，可支撑基于风险的空间分层管理（可持续发展目标6与13，Sustainable Development Goals, SDG）。